Protecting the World’s Oceans

Using “big data” to evaluate MPA e� ectiveness – the case of reefs in EU

Matthew N. K. Smith, Oceana Europe, PhD Columbia University

IntroductionThe study looked into the fi shing activity across marine Nature 2000 sites having a focus on variety of protected features protected by the Habitats Directive1. Signifi cant incompatible fi shing activity was observed inside Marine Protected Areas (MPAs) across all of the European Union (EU), particularly within reef and sandbank habitats types (codes 1170 and 1110). This poster summarizes the methods used and the main results for reefs.

This study breaks new ground by applying a comprehensive data-driven approach, extracting fi gures that offer insight into an estimate of the actual conservation status of Natura 2000 as a whole through the analysis of bottom-fi shing activities. The results demonstrate that substantive conclusions may be reached using algorithmic methods, in such a way as to be useful for lawmakers in deciding how to approach marine protection.

MethodsFishing activityFishing activity across EU waters was identifi ed using machine learning algorithms run on AIS (Automatic Identifi cation System) data from the Global Fishing Watch (GFW, globalfi shingwatch.org). The study compiled every fi shing activity of every vessel within N2000 areas for the entire year of 2017. The fi shing data was broken down by the type of fi shing gear used by each vessel and mapped across the Natura 2000 sites designated to protect specifi c habitats and species whose conservation are likely incompatible with those fi shing gears. To ensure that the analysis identifi es only fi shing activity having an impact on listed habitats, only gears with demonstrated potential impact were considered (Figure 12).

The fi shing data was gathered from satellite monitoring of vessels via AIS, matched against fl eet register databases, and compiled into tracks that encode each vessel’s course. This study used Google BigQuery to search GFW’s database for fi shing events by vessels on the register, fi shing in EU waters, in 2017. The key pieces of information extracted were the coordinates and the fi shing time, to be summed to determine the total number of hours spent fi shing inside the MPAs. There are known limitations to AIS data, however cross-checking with the EU fl eet register helps ensure completeness, mitigates false labeling and guarantees that this study cannot over-report the fi shing pressure.

To predict whether a vessel is fi shing at any given moment the GFW employs a machine learning algorithm that has been trained on a dataset that was hand-labeled as fi shing or non-fi shing by the GFW and researchers at Dalhousie University. Figure 2 shows that vessels moving very slowly or very quickly are less likely to be fi shing, while a ”sweet spot” in the middle is conducive to fi shing. Finally a logistic regression was applied using the Python package sklearn. Figure 3 demonstrates that the model is robust, accurately and precisely predicting whether a subset of trawlers are fi shing. The output of the model is a categorization of all fi shing events recorded by the GFW as fi shing or not-fi shing.

Results and discussion – case reefsThe study found signifi cant fi shing pressure on EU MPAs designated for marine habitat protection, measured as the average density of fi shing activity. Table 1 shows the largest fi shing countries by fi shing hours in 2017. The results showed that many MPAs experience fi shing pressure even far exceeding these averages for unprotected waters. The challenge in interpreting these results lies in understanding how much fi shing activity is “acceptable” for a given habitat. In this study only gear-habitat pairs with probable vulnerability were considered and this study aims to only report impactful fi shing activity.

Tables 2 and 3 show the MPAs experiencing the greatest fi shing pressure, by weighted fi shing hours and fi shing density respectively. These two measures capture two different types of MPA being affected by fi shing pressure: large MPAs with signifi cant reef complexes and small MPAs protecting isolated individual reefs. The fact that one MPA, Roches de Penmarch (FR5302008), appears in both lists, should serve to make that site stand out as particularly pressured.

European Headquarters - SpainGran Vía 5928013 Madrid, SpainPhone: +34 911 440 880E-mail: europe@oceana.org

BelgiumRue Montoyer 391000 Brussels, BelgiumPhone: +32 (0)2 513 22 42E-mail: brussels@oceana.org

DenmarkNyhavn 161051 Copenhagen, DenmarkPhone: + 45 331 51160E-mail: balti c@oceana.org

United KingdomAudley House, 13 Palace StLondon, SW1E 5HX, UKPhone: +44 20 346 87908Email: oceanauk@oceana.org

eu.oceana.org

Table 6. Matrix summarising potential vulnerability of Natura 2000 habitats to different fishing methods.

Sand

bank

s

Posid

onia

bed

s

Estu

arie

s

Mud

flats

& sa

ndfla

ts

Coas

tal l

agoo

ns

larg

e sh

allo

w in

lets

/bay

s

Reef

s

Subm

arin

e st

ruct

ures

m

ade

by le

akin

g ga

ses

Bore

al B

altic

nar

row

in

lets

Su

bmer

ged

or p

artly

su

bmer

ged

sea

cave

s

GEAR CLASS GEAR GROUP GEAR TYPE dredges DREDGES boat dredge

mechanised/suction dredge

trawls BOTTOM TRAWLS bottom otter trawl multi-rig otter trawl bottom pair trawl beam trawl

PELAGIC TRAWLS midwater otter trawl midwater pair trawl

RODS & LINES hand & pole lines hooks & lines trolling lines

LONGLINES drifting longlines set longlines

traps TRAPS pots & traps fyke nets

stationary uncovered pound nets

nets NETS trammel nets set gillnet drift net

enies esrup senies

SURROUNDING NETS lampara nets SEINES fly shooting seine

anchored seine pair seine beach & boat seine

GEAR CLASS FISHERY GEAR TYPE aquaculture shellfish rafts & longlines laid - intertidal & subtidal marine finfish cages

Probable vulnerability Possible vulnerability Unlikely vulnerability Limited information

Figure 1

1 Council Directive 92/43/EEC of 21 May 1992 on the conservation of natural habitats and of wild fauna and fl ora2 http://ec.europa.eu/environment/nature/natura2000/marine/docs/Fisheries%20interactions.pdf

Figure 2

50000

40000

30000

20000

10000

0

50000

0.0 0.2 0.4 0.6 0.8 10Measure Speed

Figure 3

0.6

0.4

0.2

0.0

0.8

0.0 0.2 0.4 0.6 0.8 10

10

Key � ndings of this study for the reefs include:

− Reefs are the most at-risk habitat across the EU. The incompatible � shing pressure on EU reefs is pervasive and intense. There are 1101 MPAs containing reef habitats in the Natura 2000 network. In 2017, incompatible � shing occurred within 242 of these, corresponding to 22% of N2000 designated for reef protection. Of these, 113 experienced more than 10 hours of � shing expected over reef habitats, 47 experienced more than 100 hours, and 10 experienced more than 1000 hours.

− Total of 2920 vessels employing incompatible gears engaged in � shing over MPAs protected for reef habitats. Of these, 846 � shed for more than 100 hours and 97 � shed for more than 1000 hours. Over 6% of all time spent � shing with bottom contacting gears in the EU waters takes place inside MPAs designated for reef protection.

− Both large MPAs with signi� cant reef complexes and small MPAs protecting isolated individual reefs experienced notable � shing pressure.

− Although Bay of Biscay (Figure 4 ) and the Adriatic Sea (Figure 5 ) experienced especially high incompatible � shing activity, signi� cant pressure was found throughout the N2000 network.

− It can be di� cult to evaluate exactly how much � shing pressure is required to result in habitat damage, but the intensity and consistency of the activity over MPAs designated for reef protection is extreme and serves as highly credible testimony that this habitat is in serious danger. These results are a call to action for proper environmental protection of vulnerable reef habitats in EU. The expansion of marine protected areas is nothing without proper enforcement.

Table 1. Average fi shing densities: largest fi shing countries by fi shing hours, 2017

Country Area of EEZ (km2) fh ρfh Dredge fh Dredge ρfh Trawl fh Trawl ρfh1. Italy 538216 1851822 3.44 6322 0.01 1546731 2.872. France 3122241 1507521 4.83 124220 0.40 961363 3.083. Spain 1020769 1386884 1.36 0 0.00 689597 0.684. UK 768286 1157984 1.51 84266 0.11 589668 0.775. Portugal 1721751 527728 0.31 151 0.00 156676 0.096. Netherlands 61869 412454 6.67 49564 0.80 315315 5.107. Denmark 102693 351904 3.43 2676 0.03 265676 2.598. Ireland 409929 256793 0.63 21216 0.05 129530 0.329. Greece 493708 191331 0.39 258 0.00 160142 0.32

10. Croatia 55961 184504 3.30 2222 0.04 97583 1.74

TOTAL (all EU): 6007939 8474896 1.41 299436 0.05 5385045 0.90

Table 2. Reefs: Top MPAs, Fishing Hours, 2017

MPA Code MPA Name Area (km2) Cover (km2) fh ρfh fh cw1. FR5400469 Pertuis Charentais 4560 586 56854 12.5 73112. BEMNZ0001 Vlaamse Banken 1099 506 12477 11.3 57423. FR5302008 Roches de Penmarch 457 174 9249 20.2 35154. DK00VA259 Gule Rev 471 250 5443 11.6 28915. PTCON0062 Banco Gorringe 22928 22673 2287 0.1 22616. FR5300031 Ile de Groix 283 153 3925 13.9 21237. UK0030385 Pobie Bank Reef 966 820 1938 2.0 16468. UK0030375 Lands End and Cape Bank 302 250 1699 5.6 14049. SE0520170 Kosterfjorden-Väderöfjorden 540 229 2916 5.4 1235

10. FR5300023 Archipel des Gl´enan 586 127 4900 8.4 1065

Table 3. Reefs: Top MPAs, Fishing Density, 2017

MPA Code MPA Name Area (km2) Cover (km2) fh ρfh fh cw1. IT3250047 Tegn`ue di Chioggia 27 1 1360 51.2 712. HR3000100 Otok Jabuka - podmorje 1 0.3 46 40.5 143. FR3102004 Ridens et dunes hydrauliques du d´etroit du Pas-de-Calais 682 7 16106 23.6 1614. DK00VA250 Store Middelgrund 21 4 499 23.4 825. FR5302008 Roches de Penmarch 457 174 9249 20.2 3515

Fishing inside Natura 2000A Python script was written to identify every fi shing event and check its location. If the coordinates fell inside an MPA, that fi shing event was labeled accordingly. Lacking information on the precise geographic location and extent of every habitat, approximations had to be made to estimate the likely activity experienced by a given habitat inside a given MPA. The following parameters were used as meaningful descriptors of fi shing pressure:

- Fishing hours- Cover-weighted fi shing hours: Fishing hours weighted to value MPAs with habitats that are

dominant in that MPA.- Average fi shing density: Total hours fi shed inside an MPA divided by the area of the MPA.

Though high values of cover-weighted fi shing hours or average fi shing density do not conclusively demonstrate that vessels were caught operating on top of the protected habitat, they indicate important patterns of fi shing pressure that imply impact on the protected habitat.

Figure 4 Figure 5

An example feature used to train the machine learning algorithm; the average speed of the vessel over a 6-hour window. Orange represents non-fi shing events, blue represents fi shing-events. A substantial difference in distribution is observed, making this feature a powerful predictor of fi shing activity.

This fi gure shows the ROC curve for the logistic regression model applied to trawlers. It demonstrates that the model is performing well. A high rate of true positives can be achieved along with a low rate of false positives

Lophelia pertusa © OCEANA.

Trawler © OCEANA / Juan Cuetos